Investigation, not yet a proposed fix
Profile of `da.stack().unstack()` on a 200×200×20 dataset shows:
```
48.09% 854 _unstack_once xarray/core/variable.py
12.61% 224 remove_unused_levels pandas/core/indexes/multi.py
3.89% 69 _engine pandas/core/indexes/multi.py
```
Nearly half the wall time is inside xarray's own `_unstack_once`. Need to read the code to determine:
- Is this fundamental numpy reshape cost (then we're done, can't optimize)?
- Is it allocation overhead from intermediate arrays?
- Is it Python-level index manipulation that could be vectorized?
Acceptance
Either: a proposed concrete fix with measured speedup, or a documented conclusion that this is irreducible.
[This is Claude Code on behalf of Felix Bumann]
Investigation, not yet a proposed fix
Profile of `da.stack().unstack()` on a 200×200×20 dataset shows:
```
48.09% 854 _unstack_once xarray/core/variable.py
12.61% 224 remove_unused_levels pandas/core/indexes/multi.py
3.89% 69 _engine pandas/core/indexes/multi.py
```
Nearly half the wall time is inside xarray's own `_unstack_once`. Need to read the code to determine:
Acceptance
Either: a proposed concrete fix with measured speedup, or a documented conclusion that this is irreducible.
[This is Claude Code on behalf of Felix Bumann]